Spatial navigation is normally yoked to the motor actions responsible for travel. Neural populations in CA1 exert influence on navigational outcomes, but this coupling makes it challenging to isolate abstract representations of travel from concrete motor actions. In this work, we present an experimental paradigm that uses a hippocampal closed-loop Brain-Machine Interface (BMI) to decouple running locomotion from the cognitive map. This BMI associates neural activity with locations on a 1D Virtual Reality (VR) track, and then translates those locations into contextually appropriate actions (Fig. a). We initially train a neural decoder on traversals of the track made using a physical controller. Despite high offline accuracy, we observe initially poor closed-loop BMI performance. By re-training a decoder on the attempted BMI traversals (Fig. b), we produce a second-generation BMI decoder that achieves high accuracy in the closed-loop setting (Fig. c). The change in controller modality, even with identical external stimuli, triggers a substantial change in the representation of the environment (Fig. d): only the second-generation BMI can successfully recover the trajectory typical of the physical controller (Fig. e, f) while decoupling motor locomotion from VR velocity (Fig. g). Interestingly, BMI usage alters the baseline representation of the environment, resulting in context superposition (Fig. d, h)—a phenomenon not observed when we remove the neural population’s influence over travel by replacing the BMI with an ‘autopilot’ controller (Fig. i). These findings suggest that influence over outcomes, whether mediated by motor actions or otherwise, is embedded in cognitive maps.